1 | /* $Id: tstRTArmv8.cpp 106061 2024-09-16 14:03:52Z vboxsync $ */
|
---|
2 | /** @file
|
---|
3 | * IPRT Testcase - armv8.h inline functions.
|
---|
4 | */
|
---|
5 |
|
---|
6 | /*
|
---|
7 | * Copyright (C) 2023-2024 Oracle and/or its affiliates.
|
---|
8 | *
|
---|
9 | * This file is part of VirtualBox base platform packages, as
|
---|
10 | * available from https://www.virtualbox.org.
|
---|
11 | *
|
---|
12 | * This program is free software; you can redistribute it and/or
|
---|
13 | * modify it under the terms of the GNU General Public License
|
---|
14 | * as published by the Free Software Foundation, in version 3 of the
|
---|
15 | * License.
|
---|
16 | *
|
---|
17 | * This program is distributed in the hope that it will be useful, but
|
---|
18 | * WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
20 | * General Public License for more details.
|
---|
21 | *
|
---|
22 | * You should have received a copy of the GNU General Public License
|
---|
23 | * along with this program; if not, see <https://www.gnu.org/licenses>.
|
---|
24 | *
|
---|
25 | * The contents of this file may alternatively be used under the terms
|
---|
26 | * of the Common Development and Distribution License Version 1.0
|
---|
27 | * (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
|
---|
28 | * in the VirtualBox distribution, in which case the provisions of the
|
---|
29 | * CDDL are applicable instead of those of the GPL.
|
---|
30 | *
|
---|
31 | * You may elect to license modified versions of this file under the
|
---|
32 | * terms and conditions of either the GPL or the CDDL or both.
|
---|
33 | *
|
---|
34 | * SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
|
---|
35 | */
|
---|
36 |
|
---|
37 |
|
---|
38 | /*********************************************************************************************************************************
|
---|
39 | * Header Files *
|
---|
40 | *********************************************************************************************************************************/
|
---|
41 | #include <iprt/asm.h> /** @todo fix me! */
|
---|
42 | #include <iprt/armv8.h>
|
---|
43 |
|
---|
44 | #include <iprt/test.h>
|
---|
45 | #include <iprt/rand.h>
|
---|
46 | #include <iprt/stream.h>
|
---|
47 | #include <iprt/sort.h>
|
---|
48 |
|
---|
49 |
|
---|
50 | /*********************************************************************************************************************************
|
---|
51 | * Defined Constants And Macros *
|
---|
52 | *********************************************************************************************************************************/
|
---|
53 | static unsigned g_cVerbosity = 0;
|
---|
54 |
|
---|
55 |
|
---|
56 | static const char *tobin(uint64_t uValue, unsigned cchWidth, char *pszBuf)
|
---|
57 | {
|
---|
58 | char *psz = pszBuf;
|
---|
59 | while (cchWidth-- > 0)
|
---|
60 | *psz++ = (uValue & RT_BIT_64(cchWidth)) ? '1' : '0';
|
---|
61 | *psz = '\0';
|
---|
62 | return pszBuf;
|
---|
63 | }
|
---|
64 |
|
---|
65 |
|
---|
66 | /** @callback_method_impl{FNRTSORTCMP} */
|
---|
67 | static DECLCALLBACK(int) cmpu32(void const *pvElement1, void const *pvElement2, void *pvUser)
|
---|
68 | {
|
---|
69 | RT_NOREF(pvUser);
|
---|
70 | if (*(uint32_t const *)pvElement1 < *(uint32_t const *)pvElement2)
|
---|
71 | return -1;
|
---|
72 | if (*(uint32_t const *)pvElement1 > *(uint32_t const *)pvElement2)
|
---|
73 | return 1;
|
---|
74 | return 0;
|
---|
75 | }
|
---|
76 |
|
---|
77 |
|
---|
78 | /** @callback_method_impl{FNRTSORTCMP} */
|
---|
79 | static DECLCALLBACK(int) cmpu64(void const *pvElement1, void const *pvElement2, void *pvUser)
|
---|
80 | {
|
---|
81 | RT_NOREF(pvUser);
|
---|
82 | if (*(uint64_t const *)pvElement1 < *(uint64_t const *)pvElement2)
|
---|
83 | return -1;
|
---|
84 | if (*(uint64_t const *)pvElement1 > *(uint64_t const *)pvElement2)
|
---|
85 | return 1;
|
---|
86 | return 0;
|
---|
87 | }
|
---|
88 |
|
---|
89 |
|
---|
90 | static unsigned BinarySearchU64(uint64_t uValue, uint64_t const *pauEntries, unsigned idxEnd)
|
---|
91 | {
|
---|
92 | unsigned idxStart = 0;
|
---|
93 | for (;;)
|
---|
94 | {
|
---|
95 | unsigned const idx = (idxEnd - idxStart) / 2 + idxStart;
|
---|
96 | uint64_t const uEntry = pauEntries[idx];
|
---|
97 | if (uValue < uEntry)
|
---|
98 | {
|
---|
99 | if (idx > idxStart)
|
---|
100 | idxEnd = idx;
|
---|
101 | else
|
---|
102 | return ~0U;
|
---|
103 | }
|
---|
104 | else if (uValue > uEntry)
|
---|
105 | {
|
---|
106 | if (idx + 1 < idxEnd)
|
---|
107 | idxStart = idx + 1;
|
---|
108 | else
|
---|
109 | return ~0U;
|
---|
110 | }
|
---|
111 | else
|
---|
112 | return idx;
|
---|
113 | }
|
---|
114 | }
|
---|
115 |
|
---|
116 |
|
---|
117 | static void tstLogicalMask32(void)
|
---|
118 | {
|
---|
119 | RTTestISub("32-bit logical masks");
|
---|
120 | static uint32_t s_auValidMasks[1312];
|
---|
121 | unsigned cValidMasks = 0;
|
---|
122 |
|
---|
123 | /* Test all legal combinations, both directions. */
|
---|
124 | char szMask[128];
|
---|
125 | char szImmR[32];
|
---|
126 | char szImmS[32];
|
---|
127 | uint32_t uImmSElmnLength = 0x3c;
|
---|
128 | for (unsigned cBitsElementLog2 = 1; cBitsElementLog2 <= 5; cBitsElementLog2++)
|
---|
129 | {
|
---|
130 | unsigned const cBitsElement = 1U << cBitsElementLog2;
|
---|
131 | for (unsigned cOneBits = 0; cOneBits < cBitsElement - 1; cOneBits++)
|
---|
132 | {
|
---|
133 | for (unsigned cRotations = 0; cRotations < cBitsElement; cRotations++)
|
---|
134 | {
|
---|
135 | uint32_t const uImmS = cOneBits | uImmSElmnLength;
|
---|
136 | uint32_t const uMask = Armv8A64ConvertImmRImmS2Mask32(uImmS, cRotations);
|
---|
137 | RTTESTI_CHECK_RETV(cValidMasks < RT_ELEMENTS(s_auValidMasks));
|
---|
138 | s_auValidMasks[cValidMasks++] = uMask;
|
---|
139 |
|
---|
140 | if (g_cVerbosity > 1)
|
---|
141 | RTPrintf("%08x %s size=%02u length=%02u rotation=%02u N=%u immr=%s imms=%s\n",
|
---|
142 | uMask, tobin(uMask, 32, szMask), cBitsElement, cOneBits, cRotations, 0,
|
---|
143 | tobin(cRotations, 6, szImmR), tobin(uImmS, 6, szImmS));
|
---|
144 |
|
---|
145 | uint32_t uImmSRev = UINT32_MAX;
|
---|
146 | uint32_t uImmRRev = UINT32_MAX;
|
---|
147 | if (!Armv8A64ConvertMask32ToImmRImmS(uMask, &uImmSRev, &uImmRRev))
|
---|
148 | RTTestIFailed("Armv8A64ConvertMask32ToImmRImmS: failed for uMask=%#x (expected immS=%#x immR=%#x)\n",
|
---|
149 | uMask, uImmS, cRotations);
|
---|
150 | else if (uImmSRev != uImmS || uImmRRev != cRotations)
|
---|
151 | RTTestIFailed("Armv8A64ConvertMask32ToImmRImmS: incorrect results for uMask=%#x: immS=%#x immR=%#x, expected %#x & %#x\n",
|
---|
152 | uMask, uImmSRev, uImmRRev, uImmS, cRotations);
|
---|
153 | }
|
---|
154 | }
|
---|
155 | uImmSElmnLength = (uImmSElmnLength << 1) & 0x3f;
|
---|
156 | }
|
---|
157 |
|
---|
158 | /* Now the other way around. */
|
---|
159 | RTSortShell(s_auValidMasks, cValidMasks, sizeof(s_auValidMasks[0]), cmpu32, NULL);
|
---|
160 | unsigned iValidMask = 0;
|
---|
161 | uint32_t uNextValidMask = s_auValidMasks[iValidMask];
|
---|
162 | uint32_t uMask = 0;
|
---|
163 | do
|
---|
164 | {
|
---|
165 | uint32_t uImmSRev = UINT32_MAX;
|
---|
166 | uint32_t uImmRRev = UINT32_MAX;
|
---|
167 | if (!Armv8A64ConvertMask32ToImmRImmS(uMask, &uImmSRev, &uImmRRev))
|
---|
168 | {
|
---|
169 | if (RT_LIKELY(uMask < uNextValidMask || iValidMask >= cValidMasks))
|
---|
170 | { }
|
---|
171 | else
|
---|
172 | {
|
---|
173 | RTTestIFailed("uMask=%#x - false negative\n", uMask);
|
---|
174 | break;
|
---|
175 | }
|
---|
176 | }
|
---|
177 | else if (RT_LIKELY(uMask == uNextValidMask && iValidMask < cValidMasks))
|
---|
178 | {
|
---|
179 | if (iValidMask + 1 < cValidMasks)
|
---|
180 | uNextValidMask = s_auValidMasks[++iValidMask];
|
---|
181 | else
|
---|
182 | {
|
---|
183 | iValidMask = cValidMasks;
|
---|
184 | uNextValidMask = UINT32_MAX;
|
---|
185 | }
|
---|
186 | }
|
---|
187 | else
|
---|
188 | {
|
---|
189 | RTTestIFailed("uMask=%#x - false positive\n", uMask);
|
---|
190 | break;
|
---|
191 | }
|
---|
192 | } while (uMask++ < UINT32_MAX);
|
---|
193 | }
|
---|
194 |
|
---|
195 |
|
---|
196 | static void tstLogicalMask64(void)
|
---|
197 | {
|
---|
198 | RTTestISub("64-bit logical masks");
|
---|
199 | static uint64_t s_auValidMasks[5376];
|
---|
200 | unsigned cValidMasks = 0;
|
---|
201 |
|
---|
202 | /* Test all legal combinations, both directions. */
|
---|
203 | char szMask[128];
|
---|
204 | char szImmR[32];
|
---|
205 | char szImmS[32];
|
---|
206 | uint32_t uImmSElmnLength = 0x3c;
|
---|
207 | for (unsigned cBitsElementLog2 = 1; cBitsElementLog2 <= 6; cBitsElementLog2++)
|
---|
208 | {
|
---|
209 | unsigned const cBitsElement = 1U << cBitsElementLog2;
|
---|
210 | for (unsigned cOneBits = 0; cOneBits < cBitsElement - 1; cOneBits++)
|
---|
211 | {
|
---|
212 | for (unsigned cRotations = 0; cRotations < cBitsElement; cRotations++)
|
---|
213 | {
|
---|
214 | uint32_t const uImmS = cOneBits | uImmSElmnLength;
|
---|
215 | uint64_t const uMask = Armv8A64ConvertImmRImmS2Mask64(uImmS, cRotations);
|
---|
216 | RTTESTI_CHECK_RETV(cValidMasks < RT_ELEMENTS(s_auValidMasks));
|
---|
217 | s_auValidMasks[cValidMasks++] = uMask;
|
---|
218 |
|
---|
219 | if (g_cVerbosity > 1)
|
---|
220 | RTPrintf("%016llx %s size=%02u length=%02u rotation=%02u N=%u immr=%s imms=%s\n",
|
---|
221 | uMask, tobin(uMask, 64, szMask), cBitsElement, cOneBits, cRotations, !!(uImmSElmnLength & 0x40),
|
---|
222 | tobin(cRotations, 6, szImmR), tobin(uImmS, 6, szImmS));
|
---|
223 |
|
---|
224 | uint32_t uImmSRev = UINT32_MAX;
|
---|
225 | uint32_t uImmRRev = UINT32_MAX;
|
---|
226 | if (!Armv8A64ConvertMask64ToImmRImmS(uMask, &uImmSRev, &uImmRRev))
|
---|
227 | RTTestIFailed("Armv8A64ConvertMask64ToImmRImmS: failed for uMask=%#llx (expected immS=%#x immR=%#x)\n",
|
---|
228 | uMask, uImmS, cRotations);
|
---|
229 | else if (uImmSRev != uImmS || uImmRRev != cRotations)
|
---|
230 | RTTestIFailed("Armv8A64ConvertMask64ToImmRImmS: incorrect results for uMask=%#llx: immS=%#x immR=%#x, expected %#x & %#x\n",
|
---|
231 | uMask, uImmSRev, uImmRRev, uImmS, cRotations);
|
---|
232 | }
|
---|
233 | }
|
---|
234 | uImmSElmnLength = (uImmSElmnLength << 1) & 0x3f;
|
---|
235 | if (cBitsElementLog2 == 5)
|
---|
236 | uImmSElmnLength = 0x40;
|
---|
237 | }
|
---|
238 |
|
---|
239 | /* Now the other way around, using random masks. */
|
---|
240 | RTSortShell(s_auValidMasks, cValidMasks, sizeof(s_auValidMasks[0]), cmpu64, NULL);
|
---|
241 |
|
---|
242 | for (unsigned iRand = 0; iRand < _32M; iRand++)
|
---|
243 | {
|
---|
244 | uint64_t const uMask = RTRandU64();
|
---|
245 | unsigned const idxMask = BinarySearchU64(uMask, s_auValidMasks, cValidMasks);
|
---|
246 | uint32_t uImmSRev = UINT32_MAX;
|
---|
247 | uint32_t uImmRRev = UINT32_MAX;
|
---|
248 | if (!Armv8A64ConvertMask64ToImmRImmS(uMask, &uImmSRev, &uImmRRev))
|
---|
249 | {
|
---|
250 | if (RT_LIKELY(idxMask == ~0U))
|
---|
251 | { }
|
---|
252 | else
|
---|
253 | {
|
---|
254 | RTTestIFailed("uMask=%#018llx - false negative\n", uMask);
|
---|
255 | break;
|
---|
256 | }
|
---|
257 | }
|
---|
258 | else if (RT_LIKELY(idxMask < cValidMasks))
|
---|
259 | { /* likely */ }
|
---|
260 | else
|
---|
261 | {
|
---|
262 | RTTestIFailed("uMask=%#018llx - false positive\n", uMask);
|
---|
263 | break;
|
---|
264 | }
|
---|
265 | }
|
---|
266 | }
|
---|
267 |
|
---|
268 |
|
---|
269 | int main()
|
---|
270 | {
|
---|
271 | RTTEST hTest;
|
---|
272 | RTEXITCODE rcExit = RTTestInitAndCreate("tstRTArmv8", &hTest);
|
---|
273 | if (rcExit != RTEXITCODE_SUCCESS)
|
---|
274 | return rcExit;
|
---|
275 |
|
---|
276 | tstLogicalMask32();
|
---|
277 | tstLogicalMask64();
|
---|
278 |
|
---|
279 | return RTTestSummaryAndDestroy(hTest);
|
---|
280 | }
|
---|
281 |
|
---|